This article has been accepted for publication in IEEE Access.

This is the author's version which has not been fully edited and
content may change prior to final publication. Citation information: DOI 10.1109/ACCESS.2023.3242917

Date of publication xxxx 00, 0000, date of current version xxxx 00, 0000.
Digital Object Identifier xx.xxxx/ACCESS.xxxx.DOI

Residual Compensation-based Extreme

Learning Machine for MIMO-NOMA
Receiver
M. REZWANUL MAHMOOD1 , MOHAMMAD ABDUL MATIN1 , (Senior Member, IEEE),
PANAGIOTIS SARIGIANNIDIS2 , (Member, IEEE), SOTIRIOS K. GOUDOS3 , (Senior
Member, IEEE), GEORGE K. KARAGIANNIDIS4 , (Fellow, IEEE)
1
Department of Electrical and Computer Engineering, North South University, Dhaka, Bangladesh
2
Department of Informatics and Telecommunications Engineering, University of Western Macedonia, Kozani, Greece
3
ELEDIA@AUTH, Department of Physics, Aristotle University of Thessaloniki, Thessaloniki 54124, Greece
4
Department of Electrical and Computer Engineering, Aristotle University of Thessaloniki, Thessaloniki, 54124, Greece
Corresponding author: Mohammad Abdul Matin (e-mail: mohammad.matin@northsouth.edu).
This work has received funding from the European Unions Horizon 2020 research and innovation programme under grant agreement No.
957406 (TERMINET).

ABSTRACT
Extreme learning machine (ELM) uses a simple machine learning (ML) architecture that allows its
implementation for smart IoT network operations. However, some inaccuracy lies between the predicted and
the actual data during the ELM training, which can be caused due to the limitation of the modeling repre-
sentation. This paper thus investigates a residual compensation-based ELM (RC-ELM) for its application in
designing a receiver for MIMO-NOMA aided IoT systems. In RC-ELM, the base ELM layers determine the
relationship between the transmitted and received data and the additional layers of the RC-ELM attempt to
compensate the error introduced during the training mechanism. The analysis of the appropriate number of
compensation layers for training error minimization is conducted on the basis of the bit error rate (BER) and
the error vector magnitude (EVM) performances of the RC-ELM training. A minimum BER improvement
of 5% for user 1 and 18% for user 2 is shown with with the aid of RC-ELM for a two-IoT user instance.
The EVM is marginally increased by 0.0008% for user 1 and 1.61% for user 2 in training stage. Besides,
the RC-ELM receiver is also compared to the minimum mean square error (MMSE), the classic ELM and
the trained multilayer perceptron (MLP) receivers in terms of BER and EVM. Both of the ELM receivers
show improved performances with respect to the other receivers.

INDEX TERMS IoT, MIMO-NOMA, machine learning, extreme learning machine, RC-ELM, multilayer
perceptron.

I. INTRODUCTION dress this issue [1]. In orthogonal multiple access (OMA),

HE deployment of 5G/B5G has created the opportunity
T to realize Internet of Things (IoT) into reality. Since
the popularity of the internet is ever increasing, the growth
of the number of communication devices is escalating expo-
nentially. Thus, multiple access technology is being brought
into attention for massive access provision to IoT devices.
However, accommodating large number of smart IoT devices
within a given bandwidth and ensuring QoS parameters
such as low latency and high throughput at the same time
can be a challenging task. Non-orthogonal multiple access
(NOMA) is seen a promising enabling technology to ad-
the users are allocated with orthogonal time/frequency re-
sources. On the contrary, in NOMA, the users are allocated
with same frequency resources at the same time. Thus,
massive accessibility with increased throughput and lower
latency is ensured. By combining NOMA with multiple-input
multiple-output (MIMO), enhanced spectral efficiency is also
achieved. MIMO-NOMA has thus received much interest in
research domain and industrial applications to enhance the
IoT network performance.
Generally, power domain NOMA is adopted for multi-
ple access provision. Multiple users with different power

VOLUME 4, 2016 1

This work is licensed under a Creative Commons Attribution 4.0 License. For more information, see https://creativecommons.org/licenses/by/4.0/
 This article has been accepted for publication in IEEE Access. This is the author's version which has not been fully edited and
content may change prior to final publication. Citation information: DOI 10.1109/ACCESS.2023.3242917
M. Rezwanul Mahmood et al.: Residual Compensation-based ELM for MIMO-NOMA Receiver
allocations are accommodated in same time and frequency
resources. Later, the signals from multiple sources are super-
imposed and transmitted to the desired destination. However,
this leads to the inter-user interference problem since the
time and frequency resources utilized by the users are non-
orthogonal, and consequently complicates the extraction of
the required information from the received signal. Thus,
several studies are being conducted on MIMO-NOMA signal
processing, which includes channel estimation, signal detec-
tion and interference cancellation operations. In [2], a zero-
forcing (ZF) beamforming is studied for multi-user down-
link MIMO-NOMA system where each cluster contains two
single-antenna user devices. Channel state information (CSI)
of each cluster-head is extracted for the ZF beamforming
design. Maximum-likelihood based receiver is utilized in [3]
for uplink MIMO-NOMA to detect quadrature phase shift
keying (QPSK) modulated symbols. A precoding scheme,
which focuses on minimization of transmission power con-
sumption, is suggested in [4] where the transmitted signal is
recovered with the help of the precoder and the minimum
mean square error (MMSE) equalizer. An iterative linear
MMSE (LMMSE) is designed in [5] multiuser detection
in MIMO-NOMA system. The mismatch between LMMSE
detection and decoding is attempted to remove in the study
for better signal recovery and achievable rate is then an-
alyzed. The performance of these approaches depends on
the accuracy of the channel estimation. Besides, the inter-
user interference mitigation approaches required in MIMO-
NOMA also relies on channel information-based techniques
[6]. However, parameters such as auto-covariance of the
channel and noise variance are challenging to obtain in
practical communication systems [7]. One possible approach
can be the adoption of machine learning (ML) techniques to
conduct such signal processing mechanisms [8].
Several studies on ML algorithm such as deep neural
network (DNN)-based approaches are conducted for MIMO-
NOMA systems to detect signals [9]–[11]. DNN based pre-
coding and decoding approaches are proposed in [9] for
MIMO-NOMA system with successive interference cancel-
lation (SIC) scheme. Uplink signal detection for MIMO-
NOMA is performed by means of convolutional neural net-
works (CNN) in [10]. The iterative soft-thresholding tech-
nique is the foundation of the adaptive DL-based receiver
architecture known as MMNet [11]. The performance of the
MMNet receiver is evaluated for massive MIMO (mMIMO)
channel with spatial correlation. However, the tuning of DNN
parameters such as weights and biases are performed by
means of iterative mechanisms, which may complicate the
implementation of such DNN algorithms. Contrary to the
tuning mechanism-based ML algorithms, extreme learning
machine (ELM) uses randomly assigned input weights and
biases (that connect between the input layer and the hidden
layer) and perform supervised learning with the help of train-
ing datasets by determining the output weights (connecting
the hidden layer and the output layer), which is responsible
for determining the desired output. In [12], [13], the ELM-
2 VOLUME 4, 2016
This work is licensed under a Creative Commons Attribution 4.0 License. For more information, see https://creativecommons.org/licenses/by/4.0/
based receiver systems are proposed which do not require
any channel information. However, due to the limitation
of modeling representation, some training inaccuracy may
be introduced during the learning stage as ML approaches,
including ELM, might not always reach the global minima
with the given ML parameters. Thus regularization tech-
nique is applied to ensure that the ML algorithms perform
well when generalized. A residual compensation-based ELM
(RC-ELM) is proposed for building the MIMO-NOMA re-
ceiver in order to reduce training error. To our knowledge,
this is the first time an RC-ELM has been researched in
relation to creating a MIMO-NOMA receiver. The following
highlights the paper’s contribution.
• An RC-ELM-based receiver is presented for a multi-
user MIMO-NOMA aided IoT system. The base ELM
layers are used to map the transmitted and received data,
whereas the additional layers are used to compensate
the error introduced during training. A bit error rate
(BER) analysis of choosing the effective number of
compensation layers to mitigate training error by means
of the RC-ELM receiver is conducted.
• The performance of the proposed ELM receiver is con-
trasted with that of the MMSE, multilayer perceptron
(MLP) and classic ELM receivers. It is demonstrated
that both of the ELM receivers provides better BER than
the MMSE and the MLP receivers.
The paper is organized as follows. Section II discusses
the MIMO-NOMA system model, which includes data trans-
mission and reception, channel estimation and signal detec-
tion with linear receiver. Section III discusses the basics of
ELM-based receiver and section IV discusses the RC-ELM
receiver. Section V presents the simulation parameters and
discussions on the results. Section VI concludes the paper.
Notations: Matrix is represented by the bold uppercase let-
ters, vector is represented by bold lowercase letters and scalar
variables are represented by small letters. The inverse, trans-
pose, conjugate transpose and Moore-Penrose inverse oper-
ations are denoted by (.)−1 , (.)T , (.)∗ and (.)† respectively.
The matrix inverse operation with regularization parameter
´ Activation function is represented by
is represented by (.).
G(.). The Frobenius norm of a matrix is denoted by ||.||F .
II. SYSTEM MODEL
A. MIMO-NOMA DATA TRANSMISSION AND RECEPTION
A visual representation of the downlink MIMO-NOMA sys-
tem is presented in Fig. 1. The source data bits for the IoT
user k are digitally modulated. Afterwards, inverse discrete
Fourier transformation (IDFT) of the data symbols of each
IoT user is carried out, followed by cyclic prefix (CP) ad-
dition. Each user is allocated with power pk and multiplied
with the respective signals. Later, the respective signals are
added. After that, the superposed data transferred to the
receiver through wireless channel [14]. After receiving the
superposed data by each IoT user, CP is removed and discrete
Fourier transformation (DFT) is carried out. The received
 This article has been accepted for publication in IEEE Access. This is the author's version which has not been fully edited and
content may change prior to final publication. Citation information: DOI 10.1109/ACCESS.2023.3242917

M. Rezwanul Mahmood et al.: Residual Compensation-based ELM for MIMO-NOMA Receiver

FIGURE 1. MIMO-NOMA system diagram. FIGURE 2. Classic ELM neural network.

DFT symbols can be mathematically expressed as shown in

(1).
Y k = F kX + Zk
Here, Y k ∈ CNR ×N is the received symbol matrix by
the user k, NR is the number of receiving antennas and
N is the number of subcarriers. F k ∈ CNR ×NT is the
channel matrix for user k, where F k = √1d H k , H k follows
k
Rayleigh distribution and dk is the distance between the
BS and the user k. X ∈ CNT ×N is the superposed signal
transmitted the BS with NT transmitting antennas and
PK by √
X = k=1 pk X k , where pk is the power allocated to the
2 weights by back-propagation algorithms, the input weights
user k. Z k is the added noise with power σZ k
.
B. CHANNEL ESTIMATION AND EQUALIZATION WITH
LINEAR RECEIVER
Channel estimation is performed by means of transmitted and
received reference symbols. If the transmitted and received
reference symbols are represented by X γ and Y γ,k , then
according to (1),
Y γ,k = F k X γ + Z k
Generally, channel is estimated in practical communica-
tion network by means of LS estimation method due to
its computational simplicity [7]. The LS estimation of the
channel F̂ k is determined as follows [15].
F̂ k,LS = (X ∗γ X γ )−1 X ∗γ Y γ,k = X †γ Y γ,k
A Bayesian-based MMSE estimation method can be used
to reduce the average mean square error (MSE) of the LS
estimation of the channel [16], [17].
∗ ∗
2
X̂ k,M M SE = (F̂ k,M M SE F̂ k,M M SE +σZ k
I)−1 F̂ k,M M SE Y k
(5)
(1)
The user with highest power can directly decode its own
signal and the user with low power needs to perform SIC
operation. Each user receives their desired data bits after
executing digital demodulation.
III. CLASSIC EXTREME LEARNING MACHINE (ELM)
FOR MIMO-NOMA SYSTEM
The classic ELM-based receiver for the MIMO-NOMA sys-
tem is presented in this section. Instead of tuning the input
as well as the biases, that connect the input layer and the
hidden layer, are randomly chosen. The output weights, that
connect the hidden layer and the output layer, are determined
in the training stage of ELM. Keeping the input weights and
biases unchanged in testing stage, the received data symbols
are processed to obtain the transmitted data.
The working principle of the ELM algorithm is demon-
strated in Fig. 2. Training of ELM algorithm is required
prior to its implementation, which is performed with the help
(2)
of transmitted-received reference symbol pairs. At first, the
received reference symbols Y γ,k are fed into the input layer,
with input neurons is equal to NR . The hidden layer output
M γ,k ∈ CL×N is determined with the help of the activation
function G(.). The output layer neuron weight parameter
β 0,k ∈ CL×NT is then determined from the transmitted ref-
erence symbols X γ and M γ,k . The following mathematical
(3)
equations state the training mechanism of ELM.
M γ,0,k = G(W k Y γ,k + bk ) (6)

β 0,k = (M ∗γ,0,k M γ,0,k + αI)−1 M ∗γ,0,k X γ

F̂ k,M M SE = σF̂2 (σF̂2 + 2
σZ )−1 F̂ k,LS (4) (7)
k,LS k,LS k = Ḿ γ,0,k X γ
Here, σF̂2 is the estimated channel variance. Since, In the equation (6) and (7), subscript (.)γ refers to terms
k,LS
MMSE approach provides better estimation of the channel, associated with reference signals. W k ∈ CL×NR and
F̂ k,M M SE is chosen for performing the equalization opera- bk ∈ CL×1 are input weight and bias parameters respec-
tion. The MMSE equalization is expressed as follows. tively which are randomly chosen and kept constant once
VOLUME 4, 2016 3

This work is licensed under a Creative Commons Attribution 4.0 License. For more information, see https://creativecommons.org/licenses/by/4.0/
 This article has been accepted for publication in IEEE Access. This is the author's version which has not been fully edited and
content may change prior to final publication. Citation information: DOI 10.1109/ACCESS.2023.3242917

M. Rezwanul Mahmood et al.: Residual Compensation-based ELM for MIMO-NOMA Receiver

FIGURE 3. RC-ELM network training stage block diagram.

assigned. The symbol α, in (7), is the regularization pa-

rameter and used to solve the least square problem β 0,k = E γ,1 = X γ − X̂ γ = X γ − β T0,k M γ,0,k (11)
argminβ0,k ||M γ,0,k β 0,k − X γ || [18].
Symbol detection is executed in testing stage. The hidden E γ,1 represents the training error. The first compensation
layer output M 0,k is computed with the received data sym- layer performs the estimation of E γ,1 according to the fol-
bols Y k as shown in (8). The detected symbols are obtained lowing equation.
from (9).
Ê γ,1 = β T1,k M γ,1,k
M 0,k = G(W k Y k + bk ) (8) = ((M ∗γ,1,k M γ,1,k + αI)−1 M ∗γ,1,k E γ,1 )T M γ,1,k
= (Ḿ γ,1,k (X γ − X̂ γ ))T M γ,1,k
X̂ k,ELM = β T0,k M 0,k (9) (12)
Here, E γ,1 is the target output in the first compensation layer.
Here, X̂ k,ELM denotes the ELM detected symbols. It The first layer prediction inaccuracy E γ,2 is then determined
is to be noted that detection by means of ELM algorithm by
takes place without determining the channel matrix directly.
It directly removes the channel effect from the received data E γ,2 = E γ,1 − Ê γ,1 . (13)
by performing online training with the help of the transmitted
and the received reference symbols. In the second compensation layer, considering E γ,2 as the
desired output, the RC-ELM training is executed as follows.
IV. RESIDUAL COMPENSATION-BASED ELM (RC-ELM)
RECEIVER FOR MIMO-NOMA Ê γ,2 = β T2,k M γ,2,k
Though the classic ELM is an effective ML algorithm for re-
= ((M ∗γ,2,k M γ,2,k + αI)−1 M ∗γ,2,k E γ,2 )T M γ,2,k
gression or classification, some error may exist in the training
of the algorithm. Suppose, X is the actual data and X̂ k,ELM = (Ḿ γ,2,k (E γ,1 − Ê γ,1 ))T M γ,2,k
is the predicted symbols with the help of the ELM receiver. = (Ḿ γ,2,k ((X γ − X̂ γ ) − Ê γ,1 ))T M γ,2,k
The estimation error can then be expressed as follows. (14)
Thus the sth compensation layer operation (where, s =
E = X − X̂ k,ELM 1, 2, . . . , S) in RC-ELM can be generally represented as
(10)
= X − β T0,k M 0,k
Ê γ,s = β Ts,k M γ,k , (15)
In order to reduce this error occurred in the training stage,
the RC-ELM receiver is presented in this paper. In RC-ELM where, β s,k = (M ∗γ,s,k M γ,s,k
+ αI)−1 M ∗γ,s,k E γ,s
=
[19], additional components (which are known as compensa- Ḿ γ,s,k E γ,s and E γ,s = E γ,s−1 −Ê γ,s−1 . The output of the
tion layers) are added to the classic ELM layers, as shown RC-ELM in the training stage would be expressed as follows.
in Fig. 3. In the input of each compensation layer, Y γ,k is
provided as input. The reference symbols Y γ,k , if were used
in (8), then the expression in (9) would provide an estimated X̂ γ,RC−ELM = X̂ γ + τ1 Ê γ,1 + τ2 Ê γ,2 + . . . + τS Ê γ,S
output X̂ γ . According to (10), the error between the actual (16a)
S
reference symbols and the ELM equalized symbols could be X
= X̂ γ + τs Ê γ,s (16b)
written as
s=1
4 VOLUME 4, 2016

This work is licensed under a Creative Commons Attribution 4.0 License. For more information, see https://creativecommons.org/licenses/by/4.0/
 This article has been accepted for publication in IEEE Access. This is the author's version which has not been fully edited and
content may change prior to final publication. Citation information: DOI 10.1109/ACCESS.2023.3242917

M. Rezwanul Mahmood et al.: Residual Compensation-based ELM for MIMO-NOMA Receiver

Algorithm 1: Summary of RC-ELM receiver design

mechanism.
TRAINING OF ELM

Input: Y γ,k , X γ
Random assignment of the input weight W k and the bias
bk
Computation of the hidden layer neuron parameter M γk
from (6)
Determination of output weight β 0,k from (7)
Computation of Ê γ,s from (15) and τs from (17) for sth
compensation layer1

TESTING OF ELM
Input: Y k , τs , and Ê γ,s
Computation of hidden layer neuron parameter M from FIGURE 4. BER vs. SNR for the classic ELM receiver as a function of
different hidden layer neurons.
(8)
Determination of X̂ k,ELM from (9)
Determination of X̂ k,RC−ELM from (18) V. SIMULATION RESULTS
The first compensation layer operation will always be performed according This section presents the simulation results for MIMO-
to (11)-(12). NOMA communication system operating at 30 GHz carrier
frequency. The user 1 and user 2 are respectively at the
distances 20 m and 10 m apart from the BS. The transmission
Here, τs represents the weights
PS of each compensation layer power is considered to be 30 dBm and the values of p1
which fulfills the criteria s=1 τs = 1. The compensation and p2 for user 1 and user 2 are respectively 0.92 and
layer weights can be calculated as follows. 0.08. Normalized 16-quadrature amplitude modulation (16-
1
QAM) is chosen in our study. Transmitted data are organized
||E γ,s −Ê γ,s ||F in physical resource block (PRB) arrangement [12]. The
τs = (17)
PS 1 transmitted and received reference symbols at one OFDM
s=1 ||E γ,s −Ê γ,s ||F
symbol time are considered for channel estimation as well as
In the testing stage, the estimated errors and the compen- the online training of both of the ELM receivers and the MLP
sation layer weights are used. After performing equalization receiver. The parameters of simulation are stated in Table 1.
according to (9), the final output by means of RC-ELM for To design both the ELM receivers, activation function tanh
the IoT user k can be represented as follows. is used [12], [13]. The input weights W and biases b follow
uniform distribution within the interval [-0.01,0.01]. The
S input layer in both the ELM receivers consists of 2 neurons
(each neuron takes the symbols received at each antenna)
X
X̂ k,RC−ELM = X̂ k,ELM + τs Ê γ,s (18)
s=1 and the output layer consists of 4 neurons (each neuron
takes the symbols transmitted from each antenna). Fig. 4
The mechanism of the RC-ELM receiver is summarized in demonstrates the BER obtained by adopting the classic ELM-
Algorithm 1. based receiver with different hidden layer neurons, which are
L1 = 2, L2 = 4 and L3 = 8. It is observed that the BER for
L2 and L3 is superior to that for L1 . Besides, the difference
in BER values for L2 and L3 is almost negligible in all SNR
TABLE 1. Simulation parameters. values. For simplicity, L2 = 4 is chosen in this study for both
of the classic ELM and the RC-ELM receivers.
Parameters Value/Technique
MIMO configuration NR × Nt 2×4 Fig. 5 demonstrates the BER obtained by the classic ELM
Number of IoT users, K 2 receiver and the RC-ELM receiver in training stage with 2, 3
Number of BS 1 and 4 compensation layers. As the regularization parameter
Modulation format 16QAM
Transmission power 30 dBm α varies from 10−2 to 10−5 , the BER performance improves
SNR (dB) 0-50 with the increase of SNR. The RC-ELM receiver is able
Number of PRB 66 to provide lower BER than the classic ELM. The RC-ELM
Number of subcarrier per PRB 12
Carrier frequency 30 GHz with 2 compensation layers shows lower BER with respect
Subcarrier spacing 120 kHz to the RC-ELM with 3 and 4 compensation layers. In Fig.
Number of OFDM symbols per slot 14 5d, the BER values of the classic ELM receiver is very
VOLUME 4, 2016 5

This work is licensed under a Creative Commons Attribution 4.0 License. For more information, see https://creativecommons.org/licenses/by/4.0/
 This article has been accepted for publication in IEEE Access. This is the author's version which has not been fully edited and
content may change prior to final publication. Citation information: DOI 10.1109/ACCESS.2023.3242917

M. Rezwanul Mahmood et al.: Residual Compensation-based ELM for MIMO-NOMA Receiver

(a) (b)

(c) (d)
−2 −3
FIGURE 5. BER for ELM and RC-ELM-based MIMO-NOMA receivers in training stage with (a) α = 10 (b) α = 10 (c) α = 10−4 (d) α = 10−5 .

similar to that of the RC-based receiver, although marginal
improvement is observed in high SNR values. For instance,
at SNR=35 dB, compared to the classic ELM receiver, the
RC-ELM provides about 5% lower BER for user 1 and
about 18% lower BER for user 2. Table 2 presents the
EVM performance (determined on the basis of the Euclidean
distance between the equalized symbols and the reference
constellation symbols) of the classic ELM and the RC-ELM
receivers with training symbols. Overall, RC-ELM with 2
compensation layers shows descent performance with respect
to the classic ELM receiver as well as the RC-ELM receivers
with 3 and 4 compensation layers. For user 1, compared to
the classic ELM, the EVM performance improvement with
RC-ELM with 2 compensation layers, within 25 − 30 dB
SNR, for α = 102 , 103 , 104 and 105 are observed to be
6 VOLUME 4, 2016
4.7%, 0.34%, 0.01% and 0.0008% respectively. For user 2,
the EVM performance, for α = 102 , 103 , 104 and 105 , is
improved by 9.2%, 13.5%, 0.28% and 1.61% within the same
SNR values. Based on the BER and the EVM performances
observed with the reference symbols, the number of compen-
sation layers in RC-ELM in testing stage is chosen to be 2.
An MLP-based receiver is also implemented in this study
for performance comparison [20]. A three hidden layer
incorporated-MLP algorithm is trained online in order to
detect the real and imaginary part of the received symbols.
For each part, the number of neurons in each layer is 4, 8
and 6 respectively. Similar to the ELM receivers, the number
of input layer and output layer is respectively chosen to be
equal to 2 and 4. Activation function tanh is also considered
for the MLP receiver at each hidden layer and the weight

This work is licensed under a Creative Commons Attribution 4.0 License. For more information, see https://creativecommons.org/licenses/by/4.0/
 This article has been accepted for publication in IEEE Access. This is the author's version which has not been fully edited and
content may change prior to final publication. Citation information: DOI 10.1109/ACCESS.2023.3242917

M. Rezwanul Mahmood et al.: Residual Compensation-based ELM for MIMO-NOMA Receiver

TABLE 2. EVM vs SNR of the classic ELM and the RC-ELM receivers in training stage for different regularization parameters.

SNR 0 5 10 15 20 25 30 35 40
Classic ELM (α = 10−2 )
User 1 45.114 29.823 20.021 14.816 13.061 12.416 12.175 12.103 12.083
User 2 127.732 123.888 111.104 102.550 98.885 95.732 93.653 93.593 91.100
RC-ELM (2 compensation layer, α = 10−2 )
User 1 44.477 28.247 17.367 10.732 8.013 7.048 7.060 7.743 8.491
User 2 126.553 123.761 107.657 98.470 91.615 82.686 79.169 84.808 88.493
RC-ELM (3 compensation layer, α = 10−2 )
User 1 44.639 28.653 18.074 11.903 9.574 8.823 8.871 9.447 10.028
User 2 127.104 122.967 108.863 100.016 94.687 87.198 87.897 91.532 89.807
RC-ELM (4 compensation layer, α = 10−2 )
User 1 44.739 28.904 18.502 12.579 10.423 9.736 9.753 10.210 10.659
User 2 127.299 123.426 109.398 101.168 95.789 89.804 90.158 92.788 90.486

Classic ELM (α = 10−3 )

User 1 47.025 29.796 18.694 10.226 6.386 3.681 2.350 1.734 1.488
User 2 130.967 119.235 111.465 101.717 86.476 77.847 74.934 74.215 72.942
RC-ELM (2 compensation layer, α = 10−3 )
User 1 45.911 28.630 18.096 10.042 6.253 3.484 2.034 1.276 0.912
User 2 131.023 119.546 112.073 99.887 85.209 71.845 63.463 54.721 43.401
RC-ELM (3 compensation layer, α = 10−3 )
User 1 46.124 28.769 18.135 10.062 6.281 3.535 2.120 1.409 1.091
User 2 130.745 119.864 111.027 100.430 85.127 74.827 68.944 61.480 55.948
RC-ELM (4 compensation layer, α = 10−3 )
User 1 46.283 28.886 18.169 10.076 6.298 3.565 2.171 1.484 1.186
User 2 130.479 119.737 111.504 100.906 85.072 75.486 72.080 66.331 57.889

Classic ELM (α = 10−4 )

User 1 46.734 27.228 15.665 8.910 4.999 2.837 1.640 0.929 0.552
User 2 128.916 122.963 108.064 94.954 76.873 64.013 45.737 31.478 1.840
RC-ELM (2 compensation layer, α = 10−4 )
User 1 46.734 27.228 15.665 8.908 4.996 2.827 1.632 0.919 0.536
User 2 128.917 122.964 108.067 94.970 76.892 63.522 45.112 31.484 1.829
RC-ELM (3 compensation layer, α = 10−4 )
User 1 46.734 27.228 15.665 8.909 4.997 2.829 1.633 0.921 0.540
User 2 128.917 122.964 108.066 94.965 76.886 63.976 45.107 31.481 1.832
RC-ELM (4 compensation layer, α = 10−4 )
User 1 46.734 27.228 15.665 8.909 4.997 2.830 1.634 0.923 0.542
User 2 128.917 122.964 108.065 94.962 76.883 64.428 45.104 31.480 1.833

Classic ELM (α = 10−5 )

User 1 44.410 26.927 15.175 8.769 5.014 2.848 1.594 0.879 0.506
User 2 128.073 121.317 110.258 94.977 82.919 66.085 54.708 40.223 14.876
RC-ELM (2 compensation layer, α = 10−5 )
User 1 44.410 26.927 15.175 8.769 5.014 2.848 1.593 0.878 0.505
User 2 128.140 121.319 109.772 95.962 82.937 65.282 53.795 37.451 12.920
RC-ELM (3 compensation layer, α = 10−5 )
User 1 44.410 26.927 15.175 8.769 5.014 2.848 1.593 0.878 0.505
User 2 128.139 121.318 110.263 95.963 83.259 65.689 53.776 37.449 12.920
RC-ELM (4 compensation layer, α = 10−5 )
User 1 44.410 26.927 15.175 8.769 5.014 2.848 1.593 0.878 0.505
User 2 128.139 121.318 110.262 95.963 83.255 66.096 53.763 37.447 12.920

parameters are updated with the learning rate of 0.005 and
momentum of 0.8. Fig. 6 presents the BER obtained by
the MMSE, the trained classic and the RC-ELMs as well
as the MLP receivers. For all the values of α, both of the
ELM receivers provide better performance than the MLP
receivers. For α = 10−5 , the BER performances by both
of the ELM receivers are almost identical. Even the ELM
receivers show improved performance against the MMSE
receiver. At the BER value of about 10−2 , about 2.5 dB and 5
dB gains are observed by the ELM receivers (with respect to
the MMSE receiver) for user 1 and user 2 respectively. Table
3 presents the EVM performance of the MMSE and the ML
VOLUME 4, 2016
algorithms (in testing stage). The ELM receivers shows better
performance for α = 10−5 .
The accuracy of the channel estimation determines the
inter-user interference mitigation and as a result, the re-
covery of the signal in NOMA-based communication net-
works. Thus, any error in the channel estimation will affect
the overall signal recovery process by means of the linear
receiver, since such receiver utilize the estimated channel
information. On the other hand, ELM receiver does not re-
quire to directly estimate channel. It rather uses the reference
symbols to perform signal detection and minimize the inter-
ference. Contrary to the complicated structure of the MLP
7

This work is licensed under a Creative Commons Attribution 4.0 License. For more information, see https://creativecommons.org/licenses/by/4.0/
 This article has been accepted for publication in IEEE Access. This is the author's version which has not been fully edited and
content may change prior to final publication. Citation information: DOI 10.1109/ACCESS.2023.3242917

M. Rezwanul Mahmood et al.: Residual Compensation-based ELM for MIMO-NOMA Receiver

(a) (b)

(c) (d)
FIGURE 6. BER for MMSE, MLP, ELM and RC-ELM-based MIMO-NOMA receivers (in testing stage with (a) α = 10−2 (b) α = 10−3 (c) α = 10−4 (d)
α = 10−5 ).

algorithm-based approach, ELM does not require tuning of
input weights and biases, which simplifies its application.
The addition of the compensation layers in ELM for RC-
ELM receiver may increase complications. However, it is
able to minimize the error caused in the training of the
ELM algorithm, which is reflected by the BER and the EVM
analyses in training stage.
shown improved performance against the MMSE and the
MLP receivers. Furthermore, it is shown that the RC-ELM
receiver training is more reliable than the conventional ELM
training. The automatic estimation of the RC-compensation
ELM’s layer count in order to enhance the performance of its
input-output mapping could be a future research target using
RC-ELM.

VI. CONCLUSION
REFERENCES
An RC-ELM based receiver is studied in this paper for
designing a receiver for the MIMO-NOMA aided IoT sys-
tem. The proposed receiver uses the classic ELM algorithm
for equalizing the symbols and the additional compensation
layers are used to minimize the errors introduced in the
classic ELM training. The BER and the EVM analyses of
both of the ELM receivers in training stage is conducted.
The performance of the RC-ELM is also compared to the
MMSE, classic ELM and MLP receivers. In terms of BER
and EVM, It is shown that both of the ELM receivers have
[1] S. R. Islam, N. Avazov, O. A. Dobre, and K.-S. Kwak, “Power-domain
non-orthogonal multiple access (NOMA) in 5G systems: Potentials and
challenges,” IEEE Communications Surveys & Tutorials, vol. 19, no. 2,
pp. 721–742, 2016.
[2] B. Kimy, S. Lim, H. Kim, S. Suh, J. Kwun, S. Choi, C. Lee, S. Lee, and
D. Hong, “Non-orthogonal multiple access in a downlink multiuser beam-
forming system,” in MILCOM 2013-2013 IEEE Military Communications
Conference. IEEE, 2013, pp. 1278–1283.
[3] J. S. Yeom, H. S. Jang, K. S. Ko, and B. C. Jung, “BER performance of up-
link NOMA with joint maximum-likelihood detector,” IEEE Transactions
on Vehicular Technology, vol. 68, no. 10, pp. 10 295–10 300, 2019.
[4] H. Wang, R. Zhang, R. Song, and S.-H. Leung, “A novel power min-

8 VOLUME 4, 2016

This work is licensed under a Creative Commons Attribution 4.0 License. For more information, see https://creativecommons.org/licenses/by/4.0/
 This article has been accepted for publication in IEEE Access. This is the author's version which has not been fully edited and
content may change prior to final publication. Citation information: DOI 10.1109/ACCESS.2023.3242917

M. Rezwanul Mahmood et al.: Residual Compensation-based ELM for MIMO-NOMA Receiver

TABLE 3. EVM performance of MMSE, MLP, classic ELM and RC-ELM receivers (in testing stage for different regularization parameters).

SNR 0 5 10 15 20 25 30 35 40
MMSE
User 1 110.562 52.055 32.148 20.966 13.71 10.67 5.382 2.711 1.595
User 2 293.536 141.744 122.59 101.899 88.089 72.782 59.499 42.971 27.917
MLP
User 1 164.653 166.454 170.906 135.689 92.724 167.884 168.269 155.838 152.643
User 2 106.954 108.428 109.209 109.37 108.12 108.907 110.219 110.849 110.053

Classic ELM (α = 10−2 )

User 1 47.068 30.679 19.157 11.993 8.005 5.928 5.136 4.835 4.728
User 2 130.228 124.465 113.331 100.481 94.405 92.786 94.189 99.026 101.332
RC-ELM (2 compensation layer, α = 10−2 )
User 1 46.939 30.683 19.091 12.015 8.086 6.083 5.311 5.023 4.898
User 2 130.35 124.404 113.178 100.588 93.955 91.211 89.881 91.655 92.664

Classic ELM (α = 10−3 )

User 1 44.602 27.325 15.671 8.928 5.073 2.904 1.684 1.099 0.809
User 2 130.255 123.489 109.456 98.478 88.551 79.345 75.446 74.869 73.636
RC-ELM (2 compensation layer, α = 10−3 )
User 1 44.603 27.324 15.675 8.927 5.076 2.908 1.700 1.121 0.838
User 2 130.350 123.627 109.780 97.803 88.320 79.510 74.205 71.641 72.264

Classic ELM (α = 10−4 )

User 1 44.837 27.134 15.511 8.895 5.041 2.850 1.600 0.898 0.516
User 2 130.845 123.838 109.345 99.084 90.607 79.279 70.281 63.313 52.863
RC-ELM (2 compensation layer, α = 10−4 )
User 1 44.837 27.133 15.512 8.899 5.043 2.850 1.601 0.898 0.516
User 2 130.569 123.713 109.487 99.132 89.783 79.359 69.518 62.693 53.099

Classic ELM (α = 10−5 )

User 1 44.850 27.137 15.611 8.872 5.056 2.913 1.687 0.957 0.549
User 2 130.999 123.093 109.548 97.125 80.412 65.184 52.280 33.963 11.131
RC-ELM (2 compensation layer, α = 10−5 )
User 1 44.850 27.137 15.611 8.874 5.063 2.919 1.688 0.956 0.549
User 2 130.952 123.091 109.551 97.196 80.249 65.535 51.971 34.316 11.309

imization precoding scheme for MIMO-NOMA uplink systems,” IEEE nications,” IEEE Transactions on Vehicular Technology, vol. 69, no. 12,
Communications Letters, vol. 22, no. 5, pp. 1106–1109, 2018. pp. 15 398–15 411, 2020.
[5] L. Liu, Y. Chi, C. Yuen, Y. L. Guan, and Y. Li, “Capacity-achieving [15] Y. S. Cho, J. Kim, W. Y. Yang, and C. G. Kang, MIMO-OFDM wireless
MIMO-NOMA: iterative LMMSE detection,” IEEE Transactions on Sig- communications with MATLAB. Singapore: John Wiley & Sons, 2010.
nal Processing, vol. 67, no. 7, pp. 1758–1773, 2019. [16] D. F. Carrera, C. Vargas-Rosales, N. M. Yungaicela-Naula, and
[6] E. N. Tominaga, O. L. López, H. Alves, R. D. Souza, and J. L. Rebelatto, L. Azpilicueta, “Comparative study of artificial neural network based
“Performance analysis of MIMO-NOMA iterative receivers for massive channel equalization methods for mmwave communications,” IEEE Ac-
connectivity,” IEEE Access, vol. 10, pp. 46 808–46 822, 2022. cess, vol. 9, pp. 41 678–41 687, 2021.
[7] J. Liu, K. Mei, X. Zhang, D. Ma, and J. Wei, “Online extreme learning [17] D. F. Carrera, C. Vargas-Rosales, L. Azpilicueta, and J. A. Galaviz-
machine-based channel estimation and equalization for OFDM systems,” Aguilar, “Comparative study of channel estimators for massive mimo 5g
IEEE Communications Letters, vol. 23, no. 7, pp. 1276–1279, 2019. nr systems,” IET Communications, vol. 14, no. 7, pp. 1175–1184, 2020.
[18] D. Gao and Q. Guo, “Extreme learning machine-based receiver for mimo
[8] M. Yao, M. Sohul, V. Marojevic, and J. H. Reed, “Artificial intelligence
led communications,” Digital Signal Processing, vol. 95, p. 102594, 2019.
defined 5G radio access networks,” IEEE Communications Magazine,
[19] J. Zhang, W. Xiao, Y. Li, and S. Zhang, “Residual compensation extreme
vol. 57, no. 3, pp. 14–20, 2019.
learning machine for regression,” Neurocomputing, vol. 311, pp. 126–136,
[9] J.-M. Kang, I.-M. Kim, and C.-J. Chun, “Deep learning-based MIMO- 2018.
NOMA with imperfect SIC decoding,” IEEE Systems Journal, vol. 14, [20] L. Noriega, “Multilayer perceptron tutorial,” School of Computing.
no. 3, pp. 3414–3417, 2019. Staffordshire University, 2005.
[10] L. Chuan, C. Qing, and L. Xianxu, “Uplink NOMA signal transmission
with convolutional neural networks approach,” Journal of Systems Engi-
neering and Electronics, vol. 31, no. 5, pp. 890–898, 2020.
[11] M. Khani, M. Alizadeh, J. Hoydis, and P. Fleming, “Adaptive neural
signal detection for massive MIMO,” IEEE Transactions on Wireless
Communications, vol. 19, no. 8, pp. 5635–5648, 2020.
[12] D. F. Carrera, D. Zabala-Blanco, C. Vargas-Rosales, and C. A. Azurdia-
Meza, “Extreme learning machine-based receiver for multi-user massive
MIMO systems,” IEEE Communications Letters, vol. 25, no. 2, pp. 484–
488, 2020.
[13] D. F. Carrera, C. Vargas-Rosales, D. Zabala-Blanco, N. M. Yungaicela-
Naula, C. A. Azurdia-Meza, M. Marey, and A. D. Firoozabadi, “Novel
multilayer extreme learning machine as a massive MIMO receiver for
millimeter wave communications,” IEEE Access, 2022.
[14] Y. Tian, G. Pan, and M.-S. Alouini, “On NOMA-based mmWave commu-

VOLUME 4, 2016 9

This work is licensed under a Creative Commons Attribution 4.0 License. For more information, see https://creativecommons.org/licenses/by/4.0/
 This article has been accepted for publication in IEEE Access. This is the author's version which has not been fully edited and
content may change prior to final publication. Citation information: DOI 10.1109/ACCESS.2023.3242917
M. Rezwanul Mahmood et al.: Residual Compensation-based ELM for MIMO-NOMA Receiver
M. REZWANUL MAHMOOD is currently work-
ing as research assistant and pursuing his M.S.
degree in Electrical and Electronic Engineering
(EEE) from North South University, Dhaka-1229,
PLACE
Bangladesh. He completed his B.S. in EEE from
PHOTO
East West University, Dhaka-1212, Bangladesh in
HERE
2017. He has published few papers in reputed
journal and international conference. His research
interests include future wireless networks, internet
of things and machine learning.
MOHAMMAD ABDUL MATIN is a Professor of
the Department of Electrical and Computer Engi-
neering at North South University (NSU), where
he has been since 2008. He received his B.Sc.
PLACE
degree in EEE from BUET (Bangladesh), M.Sc.
PHOTO
degree in Digital Communication from Loughbor-
HERE
ough University, UK and PhD in Wireless Com-
munication from Newcastle University, UK. He
has published over 120 peer-reviewed journal and
conference papers. He is the author/editor of 17
(seventeen) academic books and 21 (twenty one) book chapters. Dr. Matin
serves as a member of the editorial board for several international journals
including IEEE Communications Magazine, IET wireless Sensor Systems.
He has received a number of Prizes and Scholarships including the Best
Student Prize (Loughborough University), Commonwealth Scholarship, and
Overseas Research Scholarship (ORS) conferred by the Committee of Vice
Chancellors and Principals (CVCP) in the UK.
SOTIRIOS K. GOUDOS received the B.Sc. de-
gree in physics, the M.Sc. degree in electronics,
and the Ph.D. degree in physics from the Aristotle
University of Thessaloniki, in 1991, 1994, and
PLACE
2001, respectively, the master’s degree in infor-
PHOTO
mation systems from the University of Macedonia,
HERE
Greece, in 2005, and the Diploma degree in elec-
trical and computer engineering from the Aristotle
University of Thessaloniki, in 2011. He joined
the Department of Physics, Aristotle University
of Thessaloniki, in 2013, where he is currently an Associate Professor.
His research interests include antenna and microwave structures design,
evolutionary algorithms, wireless communications, and semantic web tech-
nologies. He is the Director of the ELEDIA@AUTH lab member of the
ELEDIA Research Center Network. Prof. Goudos is the founding Editor-in-
Chief of the Telecom open access journal (MDPI publishing). Prof. Goudos
is currently serving as Associate Editor for IEEE TRANSACTIONS ON
ANTENNAS AND PROPAGATION, IEEE ACCESS, and IEEE OPEN JOUR-
NAL OF THE COMMUNICATION SOCIETY. Prof. Goudos is currently
serving as IEEE Greece Section Secretary. He is the author of the book
"Emerging Evolutionary Algorithms for Antennas and Wireless Communi-
cations", Institution of Engineering & Technology, 2021. He is a member of
the IEEE, the IEICE, the Greek Physics Society, the Technical Chamber of
Greece, and the Greek Computer Society.
10
This work is licensed under a Creative Commons Attribution 4.0 License. For more information, see https://creativecommons.org/licenses/by/4.0/
GEORGE K. KARAGIANNIDIS was born in Pi-
thagorion, Samos Island, Greece. He received the
University Diploma (five years) and Ph.D. degrees
in electrical and computer engineering from the
PLACE
University of Patras in 1987 and 1999, respec-
PHOTO
tively. From 2000 to 2004, he was a Senior Re-
HERE
searcher at the Institute for Space Applications and
Remote Sensing, National Observatory of Athens,
Greece. In June 2004, he joined the Faculty of
the Aristotle University of Thessaloniki, Greece,
where he is currently a Professor with the Electrical and Computer En-
gineering Department and the Head of the Wireless Communications and
Information Processing (WCIP) Group. He is also a Honorary Professor at
South West Jiaotong University, Chengdu, China. His research interests are
in the broad area of digital communications systems and signal processing,
with emphasis on wireless communications, optical wireless communica-
tions, wireless power transfer and applications and communications, and
signal processing for biomedical engineering. He has been involved as
the general chair, the technical program chair, and a member of technical
program committees in several IEEE and non-IEEE conferences. In the past,
he was an editor in several IEEE journals. From 2012 to 2015, he was
the Editor-in-Chief of IEEE COMMUNICATIONS LETTERS. Currently,
he serves as the Associate Editor-in-Chief for IEEE OPEN JOURNAL OF
THE COMMUNICATIONS SOCIETY. He is one of the highly-cited authors
across all areas of electrical engineering, recognized from Clarivate Ana-
lytics as Webof-Science Highly-Cited Researcher in the seven consecutive
years 2015–2021.
PANAGIOTIS SARIGIANNIDIS is the Director
of the ITHACA lab (https://ithaca.ece.uowm.gr/),
co-founder of the 1st spinoff of the University of
Western Macedonia: MetaMind Innovations P.C.
PLACE
(https://metamind.gr), and Associate Professor in
PHOTO
the Department of Electrical and Computer Engi-
HERE
neering in the University of Western Macedonia,
Kozani, Greece. He received the B.Sc. and Ph.D.
degrees in computer science from the Aristotle
University of Thessaloniki, Thessaloniki, Greece,
in 2001 and 2007, respectively. He has published over 220 papers in
international journals, conferences and book chapters, including IEEE Com-
munications Surveys and Tutorials, IEEE Transactions on Communications,
IEEE Internet of Things, IEEE Transactions on Broadcasting, IEEE Systems
Journal, IEEE Wireless Communications Magazine, IEEE Open Journal of
the Communications Society, IEEE/OSA Journal of Lightwave Technol-
ogy, IEEE Access, and Computer Networks. His research interests include
telecommunication networks, internet of things and network security. He is
an IEEE member and participates in the Editorial Boards of various journals.
VOLUME 4, 2016